The (Yet) Problem

The Economist ran a piece on April 13, 2026 arguing that tech layoffs are not AI-caused — the headline containing the word "yet" in parentheses. The parenthetical is not a hedge. It is the acknowledgment that the argument requires a specific reading of the evidence, and that a different reading of the same evidence supports the conclusion the piece is holding at arm's length.

There are two competing theses for what is happening in the technology labor market. The cyclical correction thesis: tech overbuilt headcount during the pandemic-era growth surge, valuations collapsed with rising rates, and companies are now right-sizing to a more normal demand environment. The AI displacement thesis: the same companies are eliminating roles because AI is performing the functions those roles were hired to perform, and the right-sizing is structural rather than cyclical.

The test between these theses is not the aggregate headcount number. It is the composition of the cuts: which roles are going, which levels, and whether companies are rehiring for equivalent roles once the cycle turns.

The composition of the cuts

InformationWeek's 2026 tech layoff tracker documents the year's major cuts with function-level detail. The pattern across Q1 2026 is consistent. Atlassian cut 1,600 employees and explicitly said the cuts would "self-fund further investment in AI." Snowflake eliminated its entire technical writing and documentation department, replaced by an AI documentation system. Block attributed 40% layoffs directly to AI; its stock rose 23% on the announcement. Oracle cut 10,000-30,000 employees — senior engineers, architects, operations leaders, program managers — explicitly tied to AI data center buildout. Pendo, 10% workforce reduction, explicitly tied to "investment in AI."

These are not demand-sensitive cuts. They are companies with strong revenue growth eliminating roles because AI is performing functions those roles were hired to perform, and redirecting the freed cash flow to the AI buildout.

In 2025, AI was the documented cause of approximately 55,000 US tech layoffs per InformationWeek's tracker. The Atlanta Fed's survey of 750 corporate executives found larger companies planning AI-driven workforce reductions. Goldman Sachs found net negative AI-driven payroll flow of 16,000 per month over the past year.

What cyclical corrections look like

The cyclical correction thesis has genuine historical support. The 2022-2023 tech downturn was correctly identified as a COVID overhiring correction: companies made proportional cuts when demand didn't sustain, and rehired for similar roles when it stabilized. Those cuts were distribution-neutral — demand-sensitive functions that recover when demand does.

The current wave is different in a diagnostic way: the cuts are being made by companies with strong revenue growth. Meta's $600B AI capex commitment came alongside record profits. Oracle's cuts came alongside strong earnings. These are not companies cutting because demand is weak.

The rehiring question

The most diagnostic test between the two theses is the rehiring pattern. The Garicano, Li, and Wu framework from LSE provides the analytical lens. Their "weak-bundle" occupations — function-type jobs whose tasks can be disaggregated and assigned to AI — map directly onto what's being cut. When AI takes those tasks, humans become more efficient at what's left. Fewer of them are needed. The employment impact comes not from AI doing the whole job but from humans becoming too efficient at the remainder.

Resume.org's February 2026 survey found 21% of hiring managers citing AI as the sole reason for a pause in entry-level hiring. The Anthropic labor market paper found hiring suppression specifically in workers aged 22-25 in AI-exposed occupations. BLS employment projections show bookkeeping clerks contracting -6.2%, customer service representatives declining, data entry keyers declining — all function-type occupations at the bottom of the knowledge hierarchy.

What "yet" is doing

The Goldman Sachs scarring literature documents what the compounding looks like: a 3% real earnings cut for technology-displaced workers on return to employment, 10 percentage points slower earnings growth over the following decade, occupational downgrading that persists for years. These are not the outcomes of a cyclical correction.

The mainstream press has a structural interest in the cyclical correction thesis — it is the less alarming story, it implies market forces will eventually reabsorb displaced workers, and it avoids confronting what the displacement thesis requires confronting.

The word "yet" performs a specific function. It defers the harder question to a future in which the evidence is clearer — by which point the structural displacement will have had years more to compound.

The evidence does not sit where "yet" wants it to sit. The role composition of the cuts, the explicit company statements, the rehiring patterns, the scarring literature — it sits closer to the parenthetical's other side.